Artificial intelligence is transforming industries at breathtaking speed, but powering that transformation comes at a significant cost. A single average AI data centre consumes between 20 and 50 megawatts of electricity – enough to power tens of thousands of homes. The most advanced hyperscale facilities, purpose-built for training large language models, can demand several hundred megawatts around the clock.
To put this in perspective, training a single large AI model can consume as much energy as five cars emit over their entire lifetimes. And that is before factoring in the constant energy draw of inference – the process of actually running the model to answer queries, generate images, or analyse data. Cooling systems alone can account for 30-40% of a data centre’s total energy use.
The scale of new investment makes these numbers even more striking. The USA is currently in the midst of a historic AI infrastructure build-out, with tech giants committing hundreds of billions of dollars to new facilities. Here are some of the most significant projects underway:
- Stargate (OpenAI, Oracle and SoftBank) – The Stargate project plans to invest $500 billion over four years, with sites in Texas, New Mexico, Ohio, and Wisconsin, targeting nearly 7 gigawatts of capacity.
- Amazon Web Services – AWS is pouring $100 billion into its Generative AI Innovation Center, expanding data centre capacity across the country.
- Anthropic – Investing $50 billion in new US data centres, with flagship sites in Texas and New York expected online in 2026.
- Microsoft – Expanding aggressively across multiple states, including a major facility using a closed-loop cooling system saving over 33 million gallons of water annually.
As Bessemer Venture Partners notes, global data centre electricity consumption is projected to more than double by 2030, with US facilities soon consuming more electricity than all energy-intensive manufacturing combined. The race to build is accelerating – and so is the conversation about how to power it sustainably.